The Acquirer Divergence

How routing, retry logic, signal enrichment, and transaction data quality created a 5-to-12-point gap between the best acquirers and everyone else.

A merchant switches acquirers. Same cards. Same customers. Same products. Authorization rate jumps five points overnight.

On a billion-euro annual volume, that five-point swing is worth €50 million in recovered revenue. Not a rounding error. Not a marginal improvement. Tens of millions, sitting in the gap between one acquirer’s infrastructure and another’s.

The merchant didn’t change anything. The acquirer changed everything: where the transaction was routed, how the message was constructed, whether a soft decline got a second attempt, and what data the issuer actually received.

Last issue (sorry for missing last week, but with everyone at conferences, I wanted to I skipped it, as I think this is probably my most insightful work yet), we looked at issuers: the institution that makes the final authorization decision, but does so with incomplete information and asymmetric incentives. This issue looks at the other side of that handoff. The acquirer. The institution that determines what the issuer ever sees.

The gap between the best acquirers and the rest has grown to 5-12 percentage points for comparable transaction profiles. That gap isn’t random. It’s the result of a decade-long divergence in how acquirers think about authorization. The best treat it as a product, a competitive weapon, a reason merchants stay. The rest treat it as plumbing.

The difference is now large enough to reshape competitive position across the industry.

Let’s break it down.

How Acquiring Split in Two

For most of its history, acquiring was a volume business. Revenue is tied to transactions processed. The Merchant Service Charge, the fee merchants pay for card acceptance, contributes 60-90% of an acquirer’s total revenue.

The competitive playbook was straightforward: win merchants on price, process at scale, keep costs low. M&A was the growth engine. Ninety-one deals in 2020 across 31 countries. Investor presentations led with Total Processing Volume. The authorization rate didn’t make the slide deck.

That model worked for decades. It also meant infrastructure investment went into throughput, not intelligence. Processing margins have since been declining at -6 to 8% annually. Infrastructure providers, meaning processors and acquirers, produced the lowest returns among payments companies: -17% total shareholder return.

A different set of acquirers built their businesses on an entirely different premise.

Adyen defined authorization rate internally as “THE key competitive metric” and built its entire platform architecture around that insight. Not as an add-on. As the foundation.

Stripe built what it calls a Payments Foundation Model: self-supervised learning across tens of billions of transactions, distilling hundreds of subtle signals about each payment into a single versatile embedding. These weren’t niche players optimizing for a premium segment. They were building some of the largest acquiring platforms on earth, and they got there by treating every basis point as a product feature.

The results are visible in the numbers.

Adyen closed 2025 with €1.4 trillion in processed volume and a 53% EBITDA margin, up from 50% the year before. That margin isn’t in the transaction pipe. It’s in the intelligence layer built on top of it. Stripe processed $1.9 trillion in 2025, up 34% year-over-year, equivalent to roughly 1.6% of global GDP, and now powers more than five million businesses.

And there’s a third path emerging.

Traditional acquirers are investing seriously. Worldpay, processing 55 billion transactions a year, is building authorization intelligence across issuer data sharing, message format experimentation, and AI-driven optimization. The gap between the old model and the new one isn’t permanent. But closing it requires a fundamentally different understanding of what an acquirer does with the data flowing through its pipes.

The Four Levers That Separate Leaders from the Field

If the gap is 5-12 percentage points for comparable transaction profiles, where exactly are those points?

Having spent years working inside these systems and across 50-plus acquirers and issuers globally, I can tell you: they live in four places.

[Image 1: Gefferie’s Four Levers of Acquiring Optimization]

Routing

The routing decision, meaning which acquirer, which network, which processor endpoint handles a given transaction, is the single highest-leverage variable in authorization performance.

Local versus cross-border acquiring is the clearest example.

Issuers apply stricter risk rules to cross-border transactions. Some, particularly in France and the US, automatically decline from higher-risk geographies. A Brazilian marketplace moved from 71% to 84% authorization rates, a 13-point gain, by switching from cross-border to local acquiring. During that same process, their authorization rate in Mexico improved from 75% to 88%. Nothing else changed. Same merchant. Same customers. Different route.

For debit specifically, network selection matters more than most merchants realize.

Adyen’s Intelligent Payment Routing pilot for US debit, with eBay, Microsoft, and 24 Hour Fitness among 20-plus enterprise merchants, delivered 26% cost savings and a simultaneous 22-basis-point authorization rate improvement. Some individual merchants saw cost savings of over 50% and an auth rate improvement of over 1 percentage point.

Capgemini’s Research Institute, one of the few semi-independent voices in this space, estimates that payment orchestration’s impact on authorization rates is 2-3 percentage points through dynamic routing alone. That’s from a single lever, applied without changing anything else about the transaction.

Retry Logic

Eighty to ninety percent of all payment declines are soft declines. That means they’re temporary and potentially recoverable: insufficient funds, issuer timeouts, velocity limits. The single most common decline code across the industry is “Do Not Honor,” a catch-all that tells the merchant almost nothing about what went wrong. Close to half of all declines carry this code. In the US, the proportion is even higher.

Most acquirers retry by resubmitting the same data to the same endpoint and hoping for a different outcome. That’s not a strategy. That’s a fool’s errand.

A smart retry is an entirely different operation. It classifies every decline response against a full taxonomy: hard declines that should never be retried (stolen card, account closed), soft declines eligible for code-specific timing, and technical declines that can be retried immediately. It optimizes timing. Insufficient funds retries work best later in the day or the next day, aligned with paydays. Technical timeouts can be retried within seconds. “Do Not Honor” may succeed within minutes if the issuer was temporarily overloaded.

Smart retry also varies parameters between attempts. Adjusting amount fields. Reformatting postal codes to match issuer preferences. Routing to a different processor entirely. And it weighs every decision against cost: Visa charges $0.10- $0.15 per excess retry, while Mastercard escalates from $0.15 to $0.50.

Stripe’s approach shows how far engineering can go.

Their Adaptive Acceptance uses a TabTransformer-based deep neural network, called TabTransformer+, designed to model complex interactions among hundreds of factors that influence whether an issuer will approve a retry. It achieved 70% greater precision in identifying falsely declined transactions than the previous model, while actually reducing retry attempts by 35%. In 2024, it recovered $6 billion in falsely declined transactions, a 60% year-over-year increase. They now retrain and deploy new model versions multiple times per week.

Adyen runs 10 to 25 different versions of a given model in parallel, with at least half changing every week or two. Each version is a new idea or a retraining, A/B tested globally or for a target segment. Their own disclosure: hyperparameter tuning that took 2-3 days on CPUs runs 40 times faster on their in-house GPU infrastructure. That kind of iteration speed is itself a competitive advantage. You can’t optimize what you can’t experiment on fast enough.

The gap between basic retry and smart retry is one of the widest in payments optimization.

Signal Enrichment

The messaging standard that carries card transactions, ISO 8583, was built in 1987 for speed, not richness. It defines 128-192 data elements. A typical authorization message populates only 10-40 of those fields. That gap between what the standard allows and what actually gets sent is where approval probability quietly erodes.

The most impactful signal that almost no acquirer shares with issuers: its own fraud risk score. The acquirer has already scored the transaction as low-risk. But it doesn’t tell the issuer. The issuer makes its decision without knowing what the acquirer has already figured out.

Based on my understanding, Stripe is the furthest along here. Its Enhanced Issuer Network shares Radar fraud scores with Capital One and Discover via encrypted pathways, resulting in an 8% reduction in fraud and a 1-2% authorization rate uplift on eligible volume. No other acquirer publicly documents the practice of sharing fraud scores directly with issuers at this scale.

Worldpay is making a similar move from a different angle. Their Issuer Data Share integration with Capital One reduces false positive declines by 33%. Their Trusted Transaction product, which gives issuers a cleaner context to avoid unnecessary declines, rescued $10 million for pilot merchants in just two months. These are concrete examples of what happens when an acquirer actively helps the issuer make a better decision, rather than just throwing the transaction over the wall and hoping for the best.

The authentication layer is also an enrichment tool that most acquirers still underuse.

3DS 2.0, the protocol that supports over 150 data elements, compared to roughly 8-15 in the previous version, sends rich device and behavioral data to the issuer before the authorization decision is made. A Datos Insights/Outseer study surveying fraud executives at major financial institutions found that, in the UK, card-not-present transactions protected by 3DS achieved 90-96% authorization rates, compared with 70-75% without 3DS. That’s a 15-to-26-point differential, measured at the authorization stage, for transactions that went through the full authentication-and-authorization process.

Worldpay’s 3DS Flex Authentication Optimization Service uses AI to determine, on a per-transaction basis, whether to apply 3DS based on issuer preference, choosing the authentication path most likely to result in approval. The acquirers that treat 3DS as an optimization lever rather than a compliance checkbox are getting materially different results.

Tokenization and Message Experimentation

This is the least discussed lever and potentially one of the most accessible.

From an acquirer’s perspective, network tokenization isn’t just a security upgrade. It’s a signal quality upgrade that fundamentally changes how issuers assess risk. When a network token replaces a raw card number, it signals to the issuer that the credential is fresh, the merchant relationship is legitimate, and the card hasn’t been compromised. That signal changes the issuer’s risk calculation before any fraud model runs.

Visa reports a 4.6-percentage-point improvement in authorization rates for tokenized versus PAN-based card-not-present transactions, trending toward 6% by 2024. Mastercard reports roughly 3%. Visa has issued over 12.6 billion tokens, with a 44% year-over-year surge in 2024.

For acquirers, the strategic question isn’t whether to support tokenization. It’s how aggressively to push adoption across their merchant portfolio and how to combine token signals with their own routing and retry logic. The best acquirers use token presence as a routing signal, not just a credential format.

Beyond tokenization, experimentation with message formats is producing results that should make every payments professional take notice.

Worldpay’s experimentation with PINless debit message formats produced a 6% authorization uplift. A partnership with a major US issuer delivered a 4% uplift by adjusting the authorization request's construction. Not what was in it. How it was formatted.

The broader point is that different issuers prefer the same data presented in slightly different ways. Postal codes. Address fields. Merchant descriptors. A transaction that gets declined in one message format can be approved in another, because the issuer’s risk model interprets the signal differently. Stripe’s Adaptive Acceptance explicitly optimizes messaging and formatting alongside retry decisions. Adyen’s Smart Payment Messaging adjusts the authorization request based on the issuer. This isn’t theory. It’s production optimization happening on every transaction, and most acquirers aren’t doing it.

Why Individual Features Don’t Close the Gap

Here’s an insight I have known for a while that many acquirers and payments professionals are only now realizing.

The leading acquirers don’t have a radically different feature set. They all offer tokenization, retry logic, fraud scoring, and dynamic routing. Every serious acquirer has some version of each. The difference is in how those features connect.

At most acquirers, retry logic, routing decisions, and fraud scoring operate independently. Each feature does its job. None of them informs the others. A retry engine doesn’t know what the fraud model scored. The routing engine doesn’t know which processor had a timeout spike ten minutes ago. Fraud scoring doesn’t feed back into retry timing. Features exist. They just can’t talk to each other.

At the best acquirers, everything shares context. That’s the compounding effect.

The Stripe case study with Twilio is the most methodologically transparent example available. A/B tested across multiple major PSPs, randomized by card type, brand, country, size, and transaction type: 10% total authorization rate uplift versus the previous provider.

The breakdown:

• +5.5% from infrastructure, including local acquiring,

• +1% from Adaptive Acceptance,

• +2% from Card Account Updater,

• +1.5% from consulting recommendations.

That’s not one big feature. It’s five levers compounding.

[Image 1: Adyen's Full Funnel Payment Optimization]

Adyen’s architecture makes this visible in a way that’s worth understanding. Their NVIDIA GTC presentation revealed they run multiple ML models across the full payment funnel: checkout optimization, fraud blocking, authentication routing, authorization routing, and retry.

Each model feeds signals to the others. They tried building one global model to handle everything. It failed. What works is multiple interlinked models, each aware of the others’ decisions and the global action space. That’s why their shopper recognition rate exceeds 90% across retail merchants. When a high-value transaction arrives from an unfamiliar geography, the system already knows the customer’s pattern from another merchant. What a static rule set would decline, the platform approves.

Worldpay’s Revenue Boost follows similar integrated logic: Account Updater, Network Payment Tokens, Immediate Retries, and A/B Testing, orchestrated by a machine-learning layer that selects the optimal technique for each transaction.

[Image 2: Worldpay’s Revenue Boost Techniques]

Mastercard’s Payment Optimization Platform, announced in October 2025, is the network-level version of this same insight. POP evaluates over a trillion combinations of data elements to construct optimal authorization messages. Early pilot results with Adyen and Worldpay reported 9-15% increases in conversions. The network is building the optimization layer because the market has proven it works.

Individual features add. Orchestrated features compound. That distinction is the architectural lesson that the best acquirers proved first.

The Merchants Caught in Between

The merchants who benefit most from this divergence aren’t doing anything special. They’re just large enough to be served by the best acquirers.

Enterprise merchants processing billions choose Adyen or Stripe and get world-class optimization by default. Tier-1 merchants with sophisticated PSP partners achieve 98% SCA exemption honour rates during Europe’s PSD2 enforcement period. While smaller merchants and issuers in Southern and Eastern Europe are still recovering years later. Same regulation. Same tools available. Vastly different outcomes.

Small merchants on modern platforms benefit from the scale they don’t pay for directly. Stripe’s models, trained on $1.9 trillion in annual volume, optimize every transaction. A sole trader on Stripe gets the same Adaptive Acceptance engine that serves the Fortune 100. More than 73% of customers purchasing through Stripe Checkout have previously made payments on the Stripe network. That density of recognition is the PayFac subsidy at work: shared intelligence, distributed across the portfolio.

The merchants caught in between are feeling the gap most acutely. Processing $50-500 million annually. Too large to benefit from the platform’s aggregate ML subsidies. Too small to command dedicated enterprise attention or justify multi-acquirer orchestration. The economic threshold for multi-PSP strategies is roughly $100 million in annual volume. Below that, the complexity typically outweighs the gains. These merchants are often locked into a single traditional acquirer, and switching isn’t simple. One merchant was quoted $100,000 just to migrate card data to a different processor.

The payment orchestration market’s rapid growth, 24-26% CAGR, tells the story. Platforms like Primer and Spreedly exist because merchants needed to route around the performance gap rather than wait for their acquirer to close it. That’s market evidence of a structural problem.

The data disadvantage compounds.

Adyen’s €1.4 trillion and Stripe’s $1.9 trillion in processed volumes generate training data that mid-size acquirers can’t replicate. The models built on that data improve with scale. The acquirers already ahead keep pulling further ahead, not because the technology is secret, but because the data that feeds it accumulates faster than anyone else can keep up.

Only 64% of merchants track their false decline rates at all. Over a third don’t even know the size of the problem. That statistic is disproportionately concentrated among mid-market merchants. You can’t fix what you can’t measure, and you can’t measure what your acquirer doesn’t report.

The Correction Has Begun

Three forces are converging, at different speeds and from different directions.

From the networks. Visa’s VAMP program, launched in April 2025, replaced five previous monitoring programs and holds acquirers accountable for the fraud and dispute ratios across their entire portfolios. Penalties of $4-$8 per transaction above the threshold, $10 for merchants in the excessive tier. Thresholds tighten in January 2026 to 0.30 basis points for the above-standard designation and to 0.50 basis points for the excessive designation. For the first time, acquirers face direct financial consequences for poor data quality at the portfolio level. Not per-merchant. Portfolio-wide.

Mastercard’s Transaction Processing Excellence program imposes escalating penalties on excess retries, $0.15 up to $0.50 per violation, and introduced Merchant Advice Codes 24-30 with specific retry timing guidance for insufficient-funds declines. The network is codifying retry intelligence into its rules because too many acquirers hadn’t built their own.

From the acquirers themselves. The investment is real. Worldpay’s authorization strategy now includes dedicated auth teams, globalized auth operations, and proactive issuer advocacy. Their experimentation-driven approach, testing message formats, authentication paths, and routing strategies against live traffic, shows what it looks like when a large traditional acquirer starts treating authorization as a product discipline. Adyen’s Uplift is now available to 6,500-plus businesses, and roughly two-thirds of new merchants turn parts of it on from day one. Stripe’s Authorization Boost increases acceptance rates by 2.2% on average, with some merchants seeing up to 7%. The best are investing more, iterating faster, and pulling further ahead.

[Image 2: Stripe’s Adaptive Acceptance Improvement]

From merchants. The orchestration market’s growth signals that merchants are done waiting. Multi-PSP strategies, token vaulting, and independent routing engines. Some merchants who can afford it are working around acquirer limitations rather than advocating for change.

The question isn’t whether this gap closes. It’s whether individual acquirers close it on their own terms, or whether the networks and merchants close it for them.

What This Means for the Industry

For twenty years, the acquiring business rewarded volume above all else. The acquirers who noticed that the authorization rate was also a product and invested accordingly built an entirely different kind of business. Adyen’s 53% EBITDA margin doesn’t come from processing transactions. It comes from processing them better than anyone else and charging a premium for the difference. Stripe didn’t reach $1.9 trillion in volume by being the cheapest option. It got there by making merchants more money than they were making before.

The rest of the industry is now catching up, pushed by network penalties from above and merchant defection from below. The points are there. The tools exist. The levers are documented. The question is whether your acquirer is pulling them, and whether you’d know if they weren’t.

Issuers hold the final decision. Acquirers determine what the issuer ever sees. But both of them are increasingly operating inside an intelligence layer that neither fully controls.

In the next issue, we go inside that layer: the network scoring infrastructure that Visa and Mastercard are building, what their models actually evaluate, and what it means when decisioning logic migrates from within institutional walls to the network itself.

Thank you for reading.

---

P.S. If you are looking for a Payments Strategist/Data Scientist to help you figure out how to start leveraging data as an asset or like to educate your organization or audience through an event or webinar based on 20+ years of experience, don’t hesitate to email or DM me to set up a call and discuss how I can help.

Or if you just want to show your appreciation for my work, feel free to buy me a coffee, as it helps fuel the next editions.